Developer built 7MB AI terminal
- Developer crynta published a YouTube demo for Terax, a 7MB open-source AI terminal built with Rust, Tauri 2, React 19, and CodeMirror 6. (youtube.com) - The pitch is size and scope together — multi-tab terminal, code editor, file explorer, web preview, and AI agent loop in under 10 MB. (github.com) - It matters because Tauri’s native-webview model makes tiny desktop AI clients plausible, especially where Electron-style overhead feels wasteful. (tauri.app)
A tiny desktop app is the whole point here. Terax is an AI terminal that tries to do the modern “agentic dev tool” thing without dragging in a giant Electron shell. This week, the developer behind it — crynta — posted a demo showing a 7MB build with a terminal, editor, file explorer, web preview, and AI side panel packed into one native-feeling app. (youtube.com) (github.com) ### What actually got built? Terax is an open-source AI terminal emulator — the repo calls it an ADE, for agentic development environment — built on Tauri 2, Rust, and React 19. (tauri.app) The current feature set is not just “chat in a terminal.” It includes multi-tab terminals, a native PTY backend, an integrated code editor, file browsing, and an AI panel that can work with user-supplied API keys or local models through LM Studio. ### Why is 7MB the headline? Because most people now expect desktop AI tools to be bloated. The demo frames Terax as a reaction to that — especially to Electron-era apps that can eat hundreds of megabytes just to show text and UI chrome. (youtube.com) Crynta says Terax is roughly 50x lighter than Warp while still keeping a full AI agent loop in the app, which is the bragging-rights number that makes the project pop. ### How does it stay that small? The short answer is Tauri. Tauri apps use the operating system’s native web renderer instead of bundling a whole Chromium runtime, which is why the framework advertises app sizes that can get down to around 600KB in simple cases. (github.com) Terax is not a minimal toy, so it lands higher, but the same basic trade is doing the work here — Rust on the backend, web UI on top, no giant browser payload glued into every install. ### Is this just a terminal with a chatbot? Not really. The interesting part is the bundle of tools around the terminal. (youtube.com) Terax combines xterm.js with WebGL rendering, portable-pty for native shell access, CodeMirror 6 for editing, and AI workflows that can surface diffs and suggestions next to the code. Basically, it is trying to collapse the “terminal plus editor plus sidecar AI assistant” stack into one lightweight desktop binary. ### Why would developers care? Because a lot of AI developer tooling feels upside-down. The terminal is supposed to be the fast, invisible layer of the workflow, but newer AI wrappers often turn it into the heaviest app on the machine. (tauri.app) Terax is making the opposite argument — that local-feeling speed, small install size, OS keychain storage, and no telemetry can be product features, not just engineering trivia. ### Is this a big company launch? No — and that matters too. Right now this looks like an early open-source project from an individual developer, with a fresh GitHub repo, a brand-new demo video, and a Hacker News post used to introduce it. (github.com) That means the story is less “new market leader arrives” and more “here is a sharp prototype for a different way to build AI desktop tools.” ### So what’s the bigger pattern? The bigger pattern is that AI clients do not have to be giant general-purpose apps. If the model call is the expensive part, the desktop shell can stay lean and specialized. (github.com) Terax is a clean example of that idea — one focused binary, native integrations, bring-your-own-key support, optional local models, and just enough UI to stay useful without becoming an operating system inside an operating system. ### Bottom line? This is a small project, but it lands on a real shift. Developers are starting to treat AI desktop software less like a bloated web app and more like a compact native tool again. (github.com) If that pattern sticks, the 7MB number is not the gimmick — it is the product thesis.